README.md

metadata

dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      dtype: int64
    - name: filename
      dtype: string
  splits:
    - name: train
      num_bytes: 99150189464
      num_examples: 72000
    - name: validation
      num_bytes: 20674090404
      num_examples: 15000
    - name: test
      num_bytes: 3997875105
      num_examples: 3000
  download_size: 83930325553
  dataset_size: 123822154973
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*

---

ShinkaiGAN Image Dataset

File structure

├── train/ 
│   ├── real/ 
│   ├── anime/ 
│   └── shinkai/ 
├── val/ 
│   ├── real/ 
│   ├── anime/ 
│   └── shinkai/ 
└── test/ 
    ├── real/ 
    ├── anime/ 
    └── shinkai/

How to access it:

Dataset["{split}"][{image index}]["{feature}"]

• split: train, test, validation
• image index: pick any interger from the amount*
• features: label, filename, image**
  

*if you want to pick train anime image then pick from 0-23999, if you want to pick train real images then 24000-47999, and so on.
**label 0 is generated anime, label 1 is real images, label 2 is makoto shinkai film
**filename is the name of the file
**image is the feature containing the images it's in PIL <PIL.PngImagePlugin.PngImageFile image mode=RGB size=1024x1024>

example:

dataset["train"][0]["image"]

Dataset Overview

Pair Dataset

• Real
• Anime

Unpair Dataset

• Shinkai

Image Amount

• Train: 24,000 images x 3 = 72000 images
• Validation: 5,000 images x 3 = 15000 images
• Test: 1,000 images x 3 = 3000 images

Image Composition

• Real: LHQ1024

• Anime:

  • Generated from the real dataset using Stable Diffusion with the following prompts:
    • Prompt: "anime artwork anime style, landscape. anime style, key visual, vibrant, studio anime, highly detailed, in makoto shinkai style"
    • Negative Prompt: "photo, deformed, black and white, realism, disfigured, low contrast, cityscape, people, meteor, sunset."
    • Parameters:
      • num_steps = 10
      • guidance_scale = 0.2
      • controlnet_conditioning_scale = 0.9
      • seed = random.randint(0000000000, 9999999999)

• Shinkai:

  • Sourced from Makoto Shinkai anime in 1080p, cropped to 1080 x 1080.
  • Titles Included:
    • Your Name (Kimi no nawa)
    • Weathering with You (Tenki no Ko)
    • Suzume
    • Garden of Words
    • 5 Centimeters per Second
    • Children Who Chase Lost Voices

Contributor

• ⭐ Dwiyasa Nakula - Sepuluh Nopember Institute of Technology (Institut Technologi Sepuluh Nopember - ITS)
• ShinkaiGAN Team - Sepuluh Nopember Institute of Technology (Institut Technologi Sepuluh Nopember - ITS)